Anthropic urges the AI industry to slow down but keeps its $517B compute spending intact

Anthropic CEO Dario Amodei published an essay on September 12 titled “We Must Pace the Frontier,” calling on the AI industry to deliberately slow down the development of advanced capabilities. The piece warns about the risks of self-improving models and rogue AI agents, proposes embedding third-party evaluators inside AI labs, and advocates for coordinated safety standards among democratic nations. What the essay does not propose: buying fewer Nvidia chips. Anthropic has committed to spending $517 billion on compute resources, targeting 14.8 gigawatts of capacity through August 2026. The company’s revenue trajectory tells a similar story, projected to leap from $9 billion to a $65 billion annual run-rate within seven months. Safety rhetoric, infrastructure reality Amodei’s essay paints a picture of an industry racing toward potentially catastrophic outcomes, citing documented cases of AI misuse. A threat-intelligence report released by Anthropic on September 11-12 detailed instances of its Claude models being used for illegal activities. Amodei’s proposals focus on capability pacing, not capacity reduction. He wants to slow down what AI models can do, not how much hardware exists t...